Re: [Qemu-devel] [PATCH 0/3] write barrier support

[Christoph Hellwig]

On Tue, May 05, 2009 at 04:53:25PM +0300, Avi Kivity wrote:
> I suggest emulating barriers for format drivers that don't support them 
> by queuing the writes in the block layer and issuing a flush_all command.

We could potentially move the thread pool in posix-aio-compat.c to
work ontop of the QEMU block API instead of the Posix API which would
take care of this.  It might also help replacing the state machines
in qcow/qcow2 with a simpler threaded model, but that would need
separate investigation.